How we automatically classify ICP fit for every lead sourced from Sales Navigator

What we set out to do

Our ICP is fairly specific: the right company size, the right industries, and the right persona. Getting all three to align on a single lead takes judgment that used to live entirely in a human brain.

The challenge: Sales Navigator searches return hundreds of raw leads that broadly match with our specified keywords while sourcing leads. Manually reviewing each one to decide whether it's a Tier 1, Tier 2, Tier 3 or a skip is not feasible as we have hundreds (sometimes even thousands) of leads that we want to prepare for our outreach. Additionally, the signals of a lead can vary heavily depending on their profile, making it difficult to use simple automation logics for classifying the leads.

The goal: automate that classification process with AI, so every lead gets a consistent ICP score before it ever lands in our CRM (HubSpot).

How we set it up

The workflow chains two Phantoms and an LLM call in between:

Step 1 — Sales Navigator Search Export: We run a Sales Navigator Search Export Phantom with our saved searches (filtered by industry, geography, keywords and seniority). Output: a CSV / list in PhantomBuster with name, job title, job description, company, LinkedIn URL, company size, etc.

Step 2 — ICP Classification via Phantom's “AI LinkedIn Profile Enricher”: When the Sales Navigator Search Export is finished, another Phantom automatically runs through the freshly sourced leads to classify the ICP (and a few other properties) for each lead. Here, we precisely describe the typical characteristics for each ICP tier we have, and try to keep the classification process as strict as possible (we need to find a balance between avoiding bad leads and having a broad enough pool of leads for our outreach, which is harder than we have thought at the beginning). We also specify which properties scraped from the LinkedIn profiles of the leads should be used for the assessment (e.g. current job title and current job description). The outcome will be an updated lead list / CSV.

Step 3 — Cleaning and Enrichment of Data: For the leads, we perform additional cleaning steps (e.g. cleaning the first and last names or finding out the Emails). This partly happens directly in PhantomBuster and partly with other connected tools, like n8n or Dropcontact.

Step 4 — Push to CRM: The classified and cleaned output gets pushed into our CRM (HubSpot) with the respective ICP tier. Within HubSpot, we have different Workflows and sequences set-up that either trigger an automated outreach or a manual / personalised outreach flow, primarily based on the ICP tier.

For about 1000 leads, the pure ICP classification would take about 40-50 minutes within PhantomBuster (fully unattended). 

What happened

Before this workflow, a rep would spend at least an hour reviewing a batch of 200 leads. Now that same batch runs automatically, and we have rich data already in PhantomBuster (no need to change platforms / tools all the time, before we can send them to the CRM).

We also stopped losing good leads to fatigue - the ones that would've been skimmed over at the bottom of a long spreadsheet. Confidence in the Tier 1 pool improved once we knew every contact there had passed the same consistent bar.

The time saving alone justified the setup within the first days. We now use this approach for over 2 years with PhantomBuster.

Tips for others

• Your ICP prompt is everything. Spend time writing precise classification criteria - vague instructions produce inconsistent tiers. We version-control our prompt the same way we do code.
• Validate on a known set first. Before trusting the automation, run it against leads you've already classified manually. The gaps will tell you exactly where to refine your prompt.
• Treat it as a draft, not a verdict. The automation handles clear-cut cases really well. Edge cases still get a human eye before outreach goes out, and sometimes we have other information or signals that the automation workflow didn't have and which would lead to a different ICP classification / outreach prioritisation (we have a four eye principle, both before sending to CRM and before starting outreach when the leads are in our CRM).